Mirrors
/
hf_text-generation-inference
mirror of https://github.com/huggingface/text-generation-inference.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Taking into account number of shards.

This commit is contained in:
Nicolas Patry 2024-11-04 11:15:47 +01:00
parent 6624cbe7f2
commit 2dad0a2030
No known key found for this signature in database
GPG Key ID: D2920555C90F704C
1 changed files with 3 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
launcher/src/main.rs
View File

@ -1481,14 +1481,15 @@ struct ComputeType {
impl ComputeType { impl ComputeType {
fn f16_flop(&self) -> Option<u64> { fn f16_flop(&self) -> Option<u64> {
match &self.card[..] { let card_flop = match &self.card[..] {
// https://www.nvidia.com/en-us/data-center/l4/ // https://www.nvidia.com/en-us/data-center/l4/
"nvidia-l4" => Some(121 * 10u64.pow(12)), "nvidia-l4" => Some(121 * 10u64.pow(12)),
card => { card => {
tracing::warn!("Unkown compute for card {card}"); tracing::warn!("Unkown compute for card {card}");
None None
} }
} };
card_flop.map(|f| f * self.count as u64)
} }
} }
@ -1795,8 +1796,6 @@ fn main() -> Result<(), LauncherError> {
None => { None => {
// TODO figure out hardware optimal value // TODO figure out hardware optimal value
let compute_type = compute_type(num_shard); let compute_type = compute_type(num_shard);
tracing::info!("Compute type {compute_type:?}");
tracing::info!("Config {config:?}");
let compute_optimal = compute_optimal(config.as_ref(), compute_type.as_ref()); let compute_optimal = compute_optimal(config.as_ref(), compute_type.as_ref());
let default = compute_optimal.unwrap_or(4096); let default = compute_optimal.unwrap_or(4096);
let max_position_embeddings = config.and_then(|c| c.max_position_embeddings); let max_position_embeddings = config.and_then(|c| c.max_position_embeddings);